Luminous Landscape Forum
Equipment & Techniques => Computers & Peripherals => Topic started by: Huib on December 25, 2013, 07:51:54 am
-
intel i7-4930K CPU @3.4Ghz
Memory 64GB
OS 8.1Pro on 2 SSD Raid 0
Scratch on a seperate SSD
Work disk 2 x SSD Raid 0
Lightroom on a seperate SSD
Store 2 x HD 3TB Raid 0
Motherboard Asus X79WS
Photoshop CC
In order to work fast I have 2 SSD (each 500GB) in Raid 0
Later I save the files on 2x HD 3TB
However, it makes no difference in time at which disk I store.
This while the write speed of the SSD is much higher.
Takes on the HD 30 seconds saving a big file then that the SSD's just as long.
I need on the HD or the SSD 30 seconds for a *tif file of 70 Mb with 17 layers. Open in Photoshop CC the file is 265 Mb
If a copy a file of 1,5 Gb and place it on a other disk it is done in a few seconds.
I have als tried to turn off the Windows write cache buffer. No difference
How can this be? My computer builder does not know. The PC is working properly.
-
As a quick test, restrict the amount of memory PS can use (to something minimal like 2GB), then run the same tests saving a file to the HDD and SSD. There's a chance that application caching is influencing the results.
I would strongly caution you against running your 'store' drives in a RAID 0 configuration. A single disk failure with RAID 0 means you lose everything. I'd also set the OS disks to RAID 1. If you still feel you need some extra speed, buy a better quality SSD (using SLC NAND), but to be honest, I don't think you're going to get any appreciable difference in speed. Once you've started the PC and programs, given you have 64GB memory, most stuff is going to be cached in memory.
Spinning rust HDDs are quite good a sequential reads.
-
With single file transfers its possible you won't 'notice' speed differences between HDD's and SSD's. If a HDD takes a second to save a 70mb file, and an SSD does it 100 times faster.. will you really notice the difference between a second and a hundredth of a second? Only if you're watching for it..
With that said, 30 seconds for a 70mb file.. something is amiss. What type of SSD and HDD's are you using? Connected to I assume SATAIII ports? ACHI or EDI? Write caches on/off? Setup can make a big difference.
I agree with the last poster, make sure there's enough RAM available (after your page file, LR cache, PS cache, and whatever else you have running) for your OS to breath. And I agree with the RAID 0 sentiments. Unless you're off-loading each file when you're done with it.. it's a big risk.
One more place I'd look.. is your RAID controller and drivers. Especially if using a RAID controller on the MB.. Try a fast single SSD on a SATAIII port set up as ACHI, write cache off.. see what happens.
-
I would be looking at the raid configuration. First of all SSDs can have little to no performance gain if properly setup going from a single good connection. 2nd of all I agree with the above you have other issue(s) if it takes that long regardless. It sounds like you have an overly complex build for whatever reason...not sure what the builder was trying to achieve there...
-
So you are taking a 265MB file that is stored in RAM, compressing it and writing the result to either a HD or a SSD, lets say writing 70MB on a regular hard drive takes 1 second (it's less) and writing to the SSD takes 0,25 seconds, the difference in "small" sizes won't be a lot and the rest of your 30 seconds is being used to compress the file. If you are into overclocking you can easily test by under and overclocking the CPU.
If you want to increase saving times you can: uncheck LZW or ZIP compression for TIFFs or switch to PSD without previews.
-
Thanks for your reactions
I tested without knowing that6 the files had a Zip compression ::). Without the compression it takes 10 seconds. But again it doesn't matter if is on the HD or SSD.
I also tried it with a uncompressed 2,8 GB tiff file. To open it (That I can see it) in Photoshop cost 30 seconds and 17 seconds to close it. No difference in time between the SSD and HD
For people who really like to see what system they build for me I have attached a hardware report. Is this report helpful for you?
This is a workstation for me. Everything is backup on Synology hardware.
I wanted to have a very fast PC because I work sometimes with very big files and it is an investment for the next 5 years.
And who knows if Canon comes finally with 40 Mp camera :)
-
Just a quick look, but maybe the raided SSDs are plugged into a 1X pcie link raid Adapter card ? That is only a 1.5 G bits/sec bandwidth shared by the two ssds and which would also be about the saturation limit for a hard disk transfer rate. That would be typical of cheap "raid" add in card. You have to use server grade add in cards with at least a 4x or preferably an 8x pcie link width or the native intel sata controller suporting sata 6 Gbits/sec.
If that is correct, you should get rid of the raid and plug the ssds into the intel sata ports. Otherwise it will require 250 $ or more to get a decent raid card.
Frank
-
The builder of the Workstation is really sure that the hardware is optimal.
I must be a Photoshop isue!! >:(
-
It looks like two of the Kingston ssds are running at 3 Gbits/sec (240,120) banwidth.
The two (raid 0) Kingston 120 ssds are running on a 6 Gb/sec
The two (raid 0) Samsung 500 evo are running at 3 Gbit/sec
The two 3TB HD (raid 0) are on a 6 Gbit/sec link - not sure they can saturate that bandwidth. Most HDs max out at no more than 150MB/sec. Even if you got 300 MB/sec, that is still less than a 3 Gbit bandwidth.
So depending on which sets you are copying between you may see restriction.
Samsung 500s on a 3Gbit link are probably throttled and should perform better on a 6 Gbit link. The HD raid may perform just as well on a 3 Gbit link, particularly if only used for storage. The two solo Kingstons may benefit from a 6 Gbit link as well, but i don't know if they can saturate it.
Given that the board does not have enough 6 Gbit SATA links, there would have to be a compromise based on usage intent unless you want to invest in a good SATA/SAS raid controller. Also, you have to start considering the other possible bandwidth limitations on the buses.
Frank
-
The builder of the Workstation is really sure that the hardware is optimal.
I must be a Photoshop isue!! >:(
Of course he is.
For comparisons sake I just saved a 5 layer 170mb file to CS6x32 CS6x64, CCx32 and CCx64 Without compression they took (timed by PS) .2 seconds no matter to a RAID1 with two 3tb HDD's or a super fast SSD. Which tells me I'm still within the size of whatever Photoshop is using of my RAM for caching purposes.
With compression it takes 3.8 secs to the HDD RAID1 no matter x32 or x64 of any version. It takes 3.4x with CC x64 to my SSD, 3.5 with CS6x64 to my SSD, and 3.6 using CCx32 and CS6x32..
If you want to put your test file up on your server (or upload it to mine) I'll run the same tests so you can compare. My system is more modest than yours as I didn't see the point of exceeding the point of marginal returns for a image processing station.. It's a Haswell 4770k, 32gb of 2133 RAM, Win7, with a RAID1 using two fast 7200rpm 3tb drives, a Vertex 4 for the OS, a Crucial C300 for 'that other' OS, a Samsung 840 pro for a work drive, and some misc standalone 3tb drives. with a GTX7702gb GPU.. ALL drives are using 6gbps SATAIII ports because I didn't skimp on the motherboard..
Really, for the life of me I can't see why your builder ran with a MB without enough SATAIII ports for at least all the SSD's if not all drives period. With all the money you threw at this he was worried about another $100? And with the most likely bottlenecks too..
-
Your builder is an idiot. Good luck ;)
-
BTW, there was a Marvell Raid controller driver update that resolved some bandwidth issues a while back. You may want to have your system builder investigate to see if that may improve the situation. I understand that it requires motherboard manufacturer support as well. Perhaps open a support ticket with ASUS if there is not a newer version already available.
Good luck,
Frank
-
Your builder is an idiot. Good luck ;)
I'm in agreement with the sentiment, though it could have been pure greed and makes me wonder where else he saved a buck (power supply, etc).. but in my older age I tend to temper the words a bit.. ::)
-
Two thoughts:
1) Is the RAID adapter for the SSDs a fast, high-end unit? Not all RAID adapters are created equal.
2) Are you saving in some sort of compressed file format? The time to do the compressing may be the culprit.
-
Look at these tips
http://helpx.adobe.com/photoshop/kb/optimize-performance-photoshop-cs4-cs5.html
Frank
-
I really have to question running all these drives as raid o. I suspect the theoretical loss in speed of a good ssd on a 6gb/s port is negligible compared to the OP's raid 0 array. In fact given the poor speeds he is producing I suspect the raid 0 is actually slower.
That's not even mentioning the folly (imo) of the ssd raid 0 boot drive array.
it all appears rather misguided to me.
-
What I'd love to see is how much real world performance is gained with al those SSDs and Raids versus say one 500GB 840 evo for Os, lightroom previews, caching and everything needed and 2x3TB raid1 for storage... Right now I'm using a 3 year old imac with a SATA2 SSD and 1TB HD and i just can't find an excuse to upgrade it yet ;D
-
Nice that you are thinking about where the problem actually lies.
I myself am not a computer expert.
All discs are in my opinion a SAT300 except the OS (C) which is a SATA600 jack stand.
If I save the tif file on the disk (The two (raid 0) Kingston 120 ssds) it also makes no difference in time.
-
I'm not a computer expert either but the 1st thing I'd do is to test your SSDs as separate drives, ie change from raid 0.
-
Nice that you are thinking about where the problem actually lies.
I myself am not a computer expert.
All discs are in my opinion a SAT300 except the OS (C) which is a SATA600 jack stand.
If I save the tif file on the disk (The two (raid 0) Kingston 120 ssds) it also makes no difference in time.
Just curious.. are you saying that while you are not a computer "expert", you still know enough to judge the help you disagree with as unhelpful? Hmm.
And what is a jack stand? And how can it be an "opinion" about what type of port you're using? Are you using a translator for a technical issue?
Here it is in black and white: Your computer "expert" who sold you the machine and I presume made money doing do, did you a disservice at a minimum. This machine needs to be reconfigured. If you are willing to do this I'm sure the good folks in this thread can help you get it right. If not, well.. imagine if your ship is stuck in the ice..
-
Hi,
It really depends on what the bottleneck is. I seldom have seen raw disk speed as a bottleneck. If it takes PS 5 seconds to create a file it matters little if the disk can write in 0.1 or 1 second.
I usually check disk bandwith using activity monitor or just using dd: time dd if=/dev/zero of=bigfile bs=1024k count=1000
bigmacpro:MasterBackup ekr$ dd if=/dev/zero of=bigfile bs=1024k count=1000
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 9.389418 secs (111676359 bytes/sec)
bigmacpro:OWCBD ekr$ sudo dd if=/dev/zero of=bigfile bs=1024k count=1000
Password:
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 3.953733 secs (265211639 bytes/sec)
First sample 4TB HGST disk
Second sample SSD
Best regards
Erik
-
@Steve Yes I sometimes use google translation for making less mistakes :-\. And I think 'SAT600 Jack stand' must be 'connection'. Sorry for my bad English.
I sent you a private message for downloading the file i used to compare the time between your and my pc. The company Ikbenstil.nl who made this PC is pretty good in The Netherlands. They make according magazines the fastes and quietest computers and they always help me immediatly if I need help. This is the fourth compter they buid for me.
Yes, of course I want to configure the PC if that helps.
But where is the bottleneck? Sorry Erik, I don't understand what you are writing but I can ask tommorow the computer builder. Testing with HD tune shows very good benchmarks for de SSD >520 MB/s and 200 MB/s for the HD
And as I wrote before, copy and paste is very fast between the drives in explorer.
-
Based on the system report, the Intel RST driver and Marvell driver for raid have been updated on the ASUS support downloads ( to version 3.8.0.1106 for Intel and 1.2.0.1039 for Marvell ). It may or may not improve your system.
Re-reading the report it appears the Raided HD are on the Marvell 6Gb/s link and running at 1.5 Gb/s which is may be about as fast as the raid 0 configured disks will transfer unless they are 10k drives. The Samsung SSD are running on the 3 Gb/s Intel PCH link. A single Samsung can easily saturate that doing sequential transfers. If it is doing random R/W transfers, it likely will not. A raided pair should be able to exceed that.
See what results you get from diskbench for each of your drives.
http://www.tweaktown.com/articles/5801/samsung-840-evo-500gb-raid-0-ssd-report/index12.html
Frank
-
@Steve Yes I sometimes use google translation for making less mistakes :-\. And I think 'SAT600 Jack stand' must be 'connection'. Sorry for my bad English.
I sent you a private message for downloading the file i used to compare the time between your and my pc. The company Ikbenstil.nl who made this PC is pretty good in The Netherlands. They make according magazines the fastes and quietest computers and they always help me immediatly if I need help. This is the fourth compter they buid for me.
Huib -
It's helpful to know English is your 2nd (of 3rd or 4th) language, it allows me to fill in and not take you literally. Thanks.\
I didn't get the file in the PM section. There was no file and no link/URL. How did you send it and in what form?
Once I get a baseline with that file I'll have an opinion on reconfiguring. At a minimum I'd change the RAID0 on your system disk to a single SSD. I don't see the speed increase for disk I/O worth the risk if one SSD or your RAID goes down, and I don't see it worth the increase of speed to the disk at the expense of taxing the CPU/RAM with the RAID controller. It just doesn't add up to sound system management IMO. Now, back before SSD's we used to do HDD's in RAID0 to gain much needed speed because HDD's were a common bottleneck, but with new SSD's this isn't nearly the problem it once was.
By freeing that SATA3/6gbps port (we call them ports, not jack stands) we now have it open for one of your SSD's currently in a SATA2/3gbps port.. but which one I don't know yet. I'd like to work with that file first.
Let me know about that file.
-
Hi,
I presume that Photoshop does some processing before saving the image. So the time it takes to save the image is processing time + disk writing time. If disk writing time is short processing time will dominate.
Best regards
Erik
But where is the bottleneck? Sorry Erik, I don't understand what you are writing but I can ask tommorow the computer builder. Testing with HD tune shows very good benchmarks for de SSD >520 MB/s and 200 MB/s for the HD
And as I wrote before, copy and paste is very fast between the drives in explorer.
-
Sequential R/W benchmarks are usefull for understanding how well the system will perform for actvities like streaming video. Random R/W is usually closer to normal use. Eric is right about if PS is doing a lot of processing before writing out to disk, there my be little difference in the overall time.
Just curious, how did the previous system perform with this PS operation ?
Frank
-
Writing
Starting Create File Bench...
Created file: H:\Tijdelijk2\DiskBench1.bin
Size: 50331648 bytes
Time: 96 ms
Transfer Rate: 500,000 MB/s
Created file: G:\tijdelijk\DiskBench2.bin
Size: 50331648 bytes
Time: 133 ms
Transfer Rate: 360,902 MB/s
@ Frank. Thanks I already upgraded all drivers.
@ Steve You write that my builder is an idiot. Please, can you explane this?
@ Erik I also presume that Photoshop does some processing before saving the image. That should means that raid0 SSD not really helps to make you PS computer faster
-
Hi,
Having an SSD as scratch drive may speed up Photoshop.
I have checked with what I have (CS5) and a 39MP TIFF at 16bits/channel. Saving it on HD was instantanious. Than I added a layer and saved it again it took perhaps 50 times longer time.
As a side not, SSDs have faster write time, but the main benefit is that they don't have a moving head. Repositioning the head takes lots of time.
Best regards
Erik
Writing
Starting Create File Bench...
Created file: H:\Tijdelijk2\DiskBench1.bin
Size: 50331648 bytes
Time: 96 ms
Transfer Rate: 500,000 MB/s
Created file: G:\tijdelijk\DiskBench2.bin
Size: 50331648 bytes
Time: 133 ms
Transfer Rate: 360,902 MB/s
@ Frank. Thanks I already upgraded all drivers.
@ Steve You write that my builder is an idiot. Please, can you explane this?
@ Erik I also presume that Photoshop does some processing before saving the image. That should means that raid0 SSD not really helps to make you PS computer faster
-
In Photoshop, saving a file is a single threaded operation. This means that it only uses one CPU core to do all the work of saving the file before it writes it to disk. This is not a hard drive vs. SSD or a disk bandwidth issue.
If you're on CS6, you can disable compression in the preferences and that should speed things up considerably. In CS5 there is a plug in available to disable compression.
-
Thank you Sheldon. I use Photoshop CC but that will be the same.
So, can I make the conclusion that fast disks is not very important for Photoshop?
-
A fast disk is important for a scratch disk, depending on how much RAM you have and how big of files you edit.
-
In Photoshop, saving a file is a single threaded operation. This means that it only uses one CPU core to do all the work of saving the file before it writes it to disk. This is not a hard drive vs. SSD or a disk bandwidth issue.
If you're on CS6, you can disable compression in the preferences and that should speed things up considerably. In CS5 there is a plug in available to disable compression.
That's not entirely correct.
I just tested a 3.5GB file saving as a TIF. Uncompressed to my SSD it uses multi cores, all under full utilization. Compressed to my SSD it uses a single pegged near 100%, the others idle. Saving to my RAID0 HDD array, both compressed and uncompressed use multi cores.
-
??? I tried it on both the SSD and HD. Both use only a single core. That is according Sheldon
-
Writing
Starting Create File Bench...
Created file: H:\Tijdelijk2\DiskBench1.bin
Size: 50331648 bytes
Time: 96 ms
Transfer Rate: 500,000 MB/s
Created file: G:\tijdelijk\DiskBench2.bin
Size: 50331648 bytes
Time: 133 ms
Transfer Rate: 360,902 MB/s
@ Frank. Thanks I already upgraded all drivers.
@ Steve You write that my builder is an idiot. Please, can you explane this?
@ Erik I also presume that Photoshop does some processing before saving the image. That should means that raid0 SSD not really helps to make you PS computer faster
Huib - No I did not. I merely agreed with the sentiment of a poster who did and stated I'd choose different words to describe why certain choices were made. As to why? How many times do you have to read the same recommended changes or finally believe them? Many have told you the RAID0 is at the very least ill advised, hooking up fast SSD's to SATA2 ports isn't far behind. With all the money thrown at this system the builder couldn't find a motherboard with an adequate number of SATA3 ports? I know they're available, but they cost a bit more. As your builder to explain that as he spent your money. I suspect a lot more, but this isn't the point. The point is to get your system working optimally.. Still looking for that file in the PM section.
-
I copied the specification of my Asus X79WS which is not the cheapest.
Intel® X79 chipset :
2 x SATA 6Gb/s ports (gray), gray
4 x SATA 3Gb/s port (black), blue
Support Raid 0, 1, 5, 10
Marvell® PCIe 9128 controller :
2 x SATA 6Gb/s ports (gray), navy blue
Support Raid 0, 1
No SATA2 ports and 4 x SATA3 ports
Do you know a motherboard with more SATA6 ports?
I sent the link for downloading the file in a private message but you probably didn't see it. Please look again.
-
I copied the specification of my Asus X79WS which is not the cheapest.
Intel® X79 chipset :
2 x SATA 6Gb/s ports (gray), gray
4 x SATA 3Gb/s port (black), blue
Support Raid 0, 1, 5, 10
Marvell® PCIe 9128 controller :
2 x SATA 6Gb/s ports (gray), navy blue
Support Raid 0, 1
No SATA2 ports and 4 x SATA3 ports
Do you know a motherboard with more SATA6 ports?
I sent the link for downloading the file in a private message but you probably didn't see it. Please look again.
This one has (6) SATA3 ports which support (2) different on (http://www.gigabyte.us/products/product-page.aspx?pid=4047#sp)board RAID's and (4) SATA2 ports that can be used with one of the two RAID's. 4 USB3.0 and 14 USB2 ports. It appears x79 boards are a bit outdated, but this would get the job done for your setup.
This one has (8) SATA3 ports (6 internal 2 external) (http://www.gigabyte.us/products/product-page.aspx?pid=4766#sp) which support (2) different RAID's and more 4 USB3.0 ports (2 internal, 2 external) and 10 USB2.0 (8 external, 2 internal)..
Gigabtyte makes several more with 8 SATA3 ports. Check out which one best fits your needs.
This one has (8) internal SATA3 ports and (4) SATA2's, with (http://www.asus.com/us/Motherboards/X79DELUXE/#specifications)(8) USB3.0 ports and (12) USB2.0 ports.. And this company (ASUS) make several more that fit your needed specs.
I found the link. It wasn't showing in the text, but when I went to Reply w/Quote it was hiding in there.
Save to SSD with no compression 7.8s and to a RAID1 (2x3tb) 8.2s and for fun to my Synology 1813+ 8.0s. Let's try it with CS6: 8.0s with SSD, 8.0s with RAID1 and 8.2s to Synology 1813+. A bit odd.. I have my memory set at 17gs with 32gb installed, so to see how much different this makes I went to 5gb used and got an 8.2 to the SSD and using 25gb 8.4s..
Most of this time is spent processing the file, only a fraction saving it because the file is relatively small. Curious what your times were? Should be a great deal faster.
As far as core usage.. all 8 threads show from 10-20% usage at the most. Nothing close to maxing out.
I should also add I haven't rebooted in days, so I'm sure leaks have slowed my system down a bit.
-
Here it is in black and white: Your computer "expert" who sold you the machine and I presume made money doing do, did you a disservice at a minimum. This machine needs to be reconfigured. If you are willing to do this I'm sure the good folks in this thread can help you get it right. If not, well.. imagine if your ship is stuck in the ice..
You are being extremely polite and diplomatic, Steve. I think many of us would apply other words to describe the "expert" who configured that Frankenstein system.
-
:-\ If anybody says these words about this system, he also have to tell what is so terrible wrong of this workstation!!!!!!!!!!!!
As far as I can see is Sheldon the only one who makes a good post / answer in this topic.
-
:-\ If anybody says these words about this system, he also have to tell what is so terrible wrong of this workstation!!!!!!!!!!!!
As far as I can see is Sheldon the only one who makes a good post / answer in this topic.
Wow.. I hope the members of this forum remember this statement the next time you ask for help.
I'm starting to understand your system builder now.. ::)
-
You are right Steve. I apologize all.
But not for the PhotoEcosse and Kaeleare who wrote very negative about the computerbuilder without writing what so wrong is. That is very easy!
-
You are right Steve. I apologize all.
But not for the PhotoEcosse and Kaeleare who wrote very negative about the computerbuilder without writing what so wrong is. That is very easy!
No problem.. we all have short memories anyway.. ;D
-
I, nor anyone ELSE without hands-on time with it, can tell you what's wrong with it. Although I suspect that list is far longer than what's right with it. You can be butt-hurt all you like because it's finally sinking in that through your ignorance you got suckered and your pricey new machine is performing like a POS. That however doesn't change the cold hard facts. I simply tell it like it is, don't like it, that's ok with me :)
-
I copied the specification of my Asus X79WS which is not the cheapest.
Intel® X79 chipset :
2 x SATA 6Gb/s ports (gray), gray
4 x SATA 3Gb/s port (black), blue
Support Raid 0, 1, 5, 10
Marvell® PCIe 9128 controller :
2 x SATA 6Gb/s ports (gray), navy blue
Support Raid 0, 1
No SATA2 ports and 4 x SATA3 ports
Do you know a motherboard with more SATA6 ports?
I sent the link for downloading the file in a private message but you probably didn't see it. Please look again.
Hello Huib.
Little correction:
2 x SATA 6Gb/s ports (gray), gray = SATA3
4 x SATA 3Gb/s port (black), blue = SATA2 ( you wrote no SATA2 ports which is wrong )
Marvell® PCIe 9128 controller :
2 x SATA 6Gb/s ports (gray), navy blue
Sata ports from this MARVELL chips are crap
This chip is connected to PCIE 2.0 1x speed which means it has only 250MBps transfer speed for write. It is less than ONE good ssd drive. If you raid them on this controller I guess it will be slower than non raid ssd. This is much much worse than SATA2 from Intel chipset.
This system builder really didn't know what was he doing.
The solution is to rearrange the drives, remove some of them from raid and plug them into different ( appropriate ) sockets.